IBM Says AI Delayed Its Software Sales, Not Killed Them. Wall Street Isn't Convinced.
IBM has spent the past week in damage control mode, asking investors to treat Q2's software slump as a temporary blip rather than a structural problem. The pitch: customers didn't stop buying software, they just blew their budgets on AI infrastructure first and will come back once the hardware spending hangover clears.
That argument became necessary after IBM's preliminary results triggered one of its nastier stock drops in recent memory. Executives used the subsequent earnings call to hammer the same basic point: deals were delayed, not dead.
CEO Arvind Krishna offered what he clearly hoped would be reassuring numbers. "The majority of what didn't happen in the second quarter was large capex deals at large clients," he said, adding that roughly a third of those missing deals had already closed in the first three weeks of Q3. He was careful not to oversell it, acknowledging this was "a good indication, not yet full evidence" that spending had merely been deferred.
Wall Street was politely unconvinced. Evercore analyst Amit Daryanani cut straight to the question that dominated the Q&A: is this demand deferred or destroyed? Krishna held the company line throughout, insisting enterprise clients had temporarily redirected cash toward AI infrastructure, not permanently away from IBM's software portfolio.
Having argued the first half of the call that AI hadn't structurally damaged its software business, IBM then pivoted to explaining how AI would become its next major software revenue stream. The logic is tidy, if slightly circular: massive enterprise investment in AI infrastructure will eventually force companies to justify that spend with actual returns, creating demand for the orchestration and data-layer tools IBM wants to sell them.
The more interesting announcement was Project Lightwell, IBM's new service targeting enterprises still running on open source software that nobody is maintaining anymore. The pitch is that AI can now find and fix vulnerabilities in aging codebases faster than human teams, particularly following the release of Anthropic's Mythos model in April, which Krishna said had "accelerated the discovery of security vulnerabilities for clients" considerably.
Lightwell is priced at $1 million per year per customer, which sounds steep until you consider the client list IBM claimed on day one: Bank of America, Citi, Goldman Sachs, JPMorgan Chase, Mastercard, Morgan Stanley, Visa, and Wells Fargo. That is either a very impressive early cohort or a sign that banks are quietly terrified about what's lurking in their legacy stacks.
Krishna called it "a multiple billion-dollar opportunity" and said IBM intended to pursue it aggressively.
The underlying dynamic here is worth noting. IBM is essentially arguing that one AI, Mythos, is surfacing security problems in old software faster than anyone can manually address them, and that a different AI, Lightwell's, can fix those problems at scale. It's a tidy business model: AI creates urgency, IBM sells the solution.
Whether Q3 closes enough of those delayed deals to vindicate the deferral story is the only question that actually matters right now. Investors will find out soon enough.